HOW DO WE EXPLAIN AND WHEN WE CAN EXPLAIN WHAT OPPORTUNITIES DOES THIS OPEN UP TO US?
Donald V Steward
© Donald V Steward 2008
How do we explain what causes a behavior we observe? For example, consider that we observe that a house has been set on fire. How do we explain what caused it? We quite naturally, but perhaps without realizing it, combine two related methods of thinking, abduction and deduction.
During the abduction phase we think of all the explanations that could cause that behavior. An explanation that would lead to that behavior will be called a plausible explanation. For any given behavior, there are usually many plausible explanations each making its own assumptions. An ember might have flown from a neighboring house on fire. The fire might have been started by an arsonist. Or it might have been started by children playing with matches.
During the deduction phase we consider each of the plausible explanations and ask what the consequences would be if that explanation were correct. If we had done our abduction correctly, one of those consequences would be the behavior we wanted to explain, e.g. the house has been set on fire. But there are other consequences that would also be true. These are the side-effects.
These side-effects raise questions as to whether they are likely to be true. Was there a neighboring house on fire? If not, we reject that plausible explanation and move on to the other plausible explanations and ask whether their assumptions and side-effects could be true. We continue asking questions as to whether the assumptions and consequences of each explanation are true until we find an explanation for which they are true. If arson, was a container of fire accelerant found? If children playing with matches, were there children present at the time the fire started?
We may have faced the same behavior before and already have some plausible explanations to offer. This is experience. But if we havenít experienced this particular behavior before or if the current situation is not explained by our experience, then by using abduction we can often put together various pieces of knowledge we do have in order to find some plausible causes for this unfamiliar behavior.
For example, we can start with our existing knowledge such as: a fire requires the proximity of a combustible material, AND a source of oxygen, AND a local source of high heat. Each of these ANDs needs to be investigated. We start by asking what could cause a local source of high heat. We continually ask Ďwhat could be the cause of that, and the cause of that, and so oní until we get to the most fundamental cause-and-effect statements. The local high heat could be caused by an electrical spark. An electrical spark could be caused by faulty wiring. Or an electrical spark could be caused by rodents gnawing on the wiring. By this means we can develop a number of plausible explanations using ANDs, ORs and NOTs.
We might use collaboration to develop a cause-and-effect model of the circumstances to be explained. A number of people would each contribute their own pieces to the puzzle. But when people collaborate, they often run into troubles that prevent them from being able to assemble their best knowledge to solve the problem.
The collaborators may not understand each other. They may not be able to express their thoughts with sufficient clarity and logic. They may have hidden motives they donít wish to reveal. Or their statements might hide assumptions that perhaps even they may be unaware of. Often the problem has so many aspects that it is beyond their unaided ability to deal with them. So they simplify the problem. But the simplification may overlook essential factors critical to solving the problem.
This leads to what we call the Explainer problem. How can we represent our thoughts clearly so they can be expressed and combined logically to solve complex problems? And can this be done with the help of a computer so we can handle more complex problems involving more considerations than we could otherwise deal with?
Letís use the same approach we just discussed to develop such a method, which we will call the Explainer. We start with a hypothesis (abduction) and analyze its consequences (deduction) to see what the consequences would be and evaluate whether what we end up with is useful.
Letís make the hypothesis that the knowledge we need can be represented by systems of cause-and-effect statements. This seems like a good choice because cause-and-effect statements can be represented using AND, OR and NOT logic that can be used to logically put together the various individual cause-and-effect statements to form a larger picture. We have found that by following the consequences of this hypothesis, there are a number of very surprising implications.
We have found that this is a fundamental problem that is the key to solving many other types of problems that we had previously thought were quite different. This may be too audacious to be believed. Surely something must be wrong? But see NOTE below.
Cause-and-effect statements can represent a wide variety of things so that the Explainer can be used to diagnose the causes of their various behaviors and misbehaviors. Cause-and-effect knowledge about how human medical systems work can be used to diagnose medical symptoms and propose treatments. It is much easier to use this method than the typical approach to developing expert systems. Cause-and-effect statements can represent the electrical system of an automobile in order to find out why the car doesnít start, or to diagnose difficulties with many other types of systems or processes.
One can determine how to change the behavior of something by asking for an explanation for that new behavior, if such a behavior is possible. For example, by describing the situation in Iraq using cause-and-effect statements, it was possible to propose actions that could lead to reducing the violence there and maybe later to show how to remove U.S. troops with the least loss of life. However, this current model raises many questions that still need to be answered by more savvy experts before an adequate model has been completed.
The Explainer can be used to help people collaborate to put together their individual contributions of knowledge to solve very complex problems. Collaborating to solve the Iraqi problem would be an example to suggest how the Explainer can be used to propose solutions for many other very complex social problems.
The Explainer can also be used to explain what may have happened in the past such as in solving crimes, reconstructing civilizations from archeological digs, or reconstructing what may have happened at an earlier time in history.
Systems can be designed with the help of the Explainer by asking it to use the capabilities of possible components described by cause-and-effect statements to Ďexplainí the requirement specifications of whatever is being designed.
The Explainer can be made to learn. When observed behaviors are explained, the resulting explanation and the behavior explained make up a cause-and-effect statement. This cause-and-effect statement can be added to its existing cause-and-effect repertoire. This makes the Explainer recursive so it can learn.
We have used the Explainer with a set of cause-and-effect statements, examples of a few of them are shown below, to offer a plausible explanation for the cause and solution of the subprime mortgage and liquidity crisis. It offers a plausible explanation of how so many subprime mortgages were sold, what caused the costs of houses to rise and then fall, why there were so many foreclosures, and why market liquidity collapsed and the market with it.
22. Bundled loans have low risk rating
Because 20. Artificially low risk assigned to bundled loans by
20. Artificially low risk assigned to bundled loans by rating companies
Because 26. Rating companies are financially supported by those
26. Rating companies are financially supported by those selling bundles
Because NOT 35 Regulators prevent rating companies from being
supported by those selling the bundles.
Most importantly, it showed that the basic cause of the crisis was that government did not assure that adequate information was available so people making transactions could properly evaluate their risks. This is a basic premise that must hold for a free market to operate properly. Most of what is being done instead is unfortunately using humungous amounts of money to resolve the symptoms without getting at the basic cause of the problem.
People were running around saying something big should be done and done quickly. And it was. But what that something was ambiguous, allowing everyone to jump in with their own self-serving proposals. The result was uncertainty and a tumultuous market. What is needed is not more money, but more information.
Building on the concepts above, there may be many other ideas about how the Explainer can be used to solve other types of problems that have not been considered here. Perhaps you can think of some. If you have some ideas, I would be interested in helping you develop them. You can reach me at Steward@Problematics.Com or by phone at (707) 226-5102.
A more formal development of this method can be found at: http://problematics.com/articles/Behavior-E.doc . A set of trial applications made using the Explainer can be found at: http://problematics.com/articles/APPLICATIONS-E.doc
NOTE: This Explainer method sounds too audacious to be believed. So one is inclined to try to show that this problem canít be solved and then dismiss it. Many mathematicians apparently think this problem falls into a class of problems known to be NP-Complete. NP-Complete problems can only be solved by enumerating and testing all possibilities, then eliminating all those that donít produce a plausible explanation, until finally just those explanations that are plausible are left. But this would be impractical for all but very small problems.
But if they were to look at this problem more carefully, they would see that provided there are no circuits in the cause-and-effect statements, there is a straight forward and efficient way of directly finding just the plausible explanations as a function of the behavior by using substitutions to eliminate the intermediate variables. And a surprisingly large number of interesting problems fall into this class.
Then if they looked yet more carefully, they would see that the problem does indeed have NP-Complete characteristics if there are circuits in the cause-and-effect statements. But they will also see that efficient solutions can still be found for many, but not necessarily all, interesting problems even when they do contain circuits.
The difficulty is to get people to look at this carefully without prematurely jumping to conclusions. But stubborn perseverance will eventually reveal a method that has a surprising number if interesting and important implications and applications.
The Explainer could be used by a medical group who keep up with the latest literature to maintain a diagnostic system sold as a subscription to physicians. It could be made available as a Wiki for volunteers to develop diagnostic packages for all sorts of things. Or it could be made available for people to collaborate in solving all types of complex problems. It could be very useful to government.
So is this audacious? Absolutely! But can it be used to find better and sometimes less deadly solutions to complex problems? Apparently so! Now we need people to use it to solve the many complex problems whose improper solutions often vex and sometimes endanger our society.